INSTITUTE OF PHYSICS PUBLISHING EUROPEAN JOURNAL OF PHYSICS

Eur. J. Phys. 25 (2004) 193–202 PII: S0143-0807(04)69172-0

The efficiency of overshot and

undershot waterwheels
M Denny
5114 Sandgate Road, Victoria, BC, V9C 3Z2, Canada

E-mail: markandjane@shaw.ca

Received 5 September 2003

Published 2 December 2003
Online at stacks.iop.org/EJP/25/193 (DOI: 10.1088/0143-0807/25/2/006)

Abstract
The waterwheel evolved over two millennia to become an efficient machine.
We analyse the physics of waterwheels, and construct simple models that show
why the two most important types had very different efficiencies. Our analysis
reveals the important design parameters, and captures the essential features of
our oldest mechanical power source.

1. Introduction and history

Physicists commonly adopt the waterwheel as an analogy for other physical systems; the flow
of water may represent the flow of electricity, or steam, with the capacity to do work. More
recently it has been adopted as a representation of the Lorenz equations, so that for certain
regions of the parameter space, this strange waterwheel behaves chaotically1 [1]. In this paper
we shall consider this oldest of machines not as an analogy,nor as a representation of some other
physical system, but shall instead investigate the historically important waterwheel designs,
to see why they developed the way they did. In particular, we shall calculate the efficiencies
of several types of waterwheel. This is an instructive application of Newton’s laws, energy
transfer, power, torque, and elementary fluid mechanics to a familiar and important machine,
which will shed light upon the design features that exercised the minds of many well known
and unknown scientists and engineers over the ages.
The waterwheel has evolved steadily since it was introduced 2000 years ago, to pump
water and mill grain. From the rather scant records of classical antiquity it is not clear where
it originated; it is clear that it spread rapidly and is described by Roman, Greek and Chinese
sources. These early machines (the ‘Greek’ or ‘Norse’ mill) had horizontal wheels, i.e. with
vertical shafts, since these are simplest and required no gearing to transmit power to the
millstone [2, 3]. There is good evidence that the familiar vertical waterwheel (with horizontal
axle) developed within the Roman Empire [4] and spread out rapidly from there. The undershot
waterwheel (so-called because the water passes underneath the axle) is described by Vitruvius
1 A chaotic waterwheel has been built at the Fachhochschule Brugg-Windisch (Switzerland). The chaotic character
of this device is described in the website http://people.web.psi.ch/gassmann/waterwheel/WaterwheelLab.html, where
an interactive simulation is available.

0143-0807/04/020193+10$30.00 © 2004 IOP Publishing Ltd Printed in the UK 193

194 M Denny

in 27 BC [3, 4]. This design was more common but less efficient than the overshot waterwheel
until the 13th century [2], though the undershot type continued to be popular thereafter, for
reasons we shall discuss below. In the 19th century it was made much more efficient, as we
shall see, because of a development in France that anticipated the successor of waterwheel
technology: hydraulic turbines.
It is hard to overstate the historical importance of waterwheels. They were the primary
source of power from antiquity until the introduction of reliable high-pressure steam engines
at the end of the 18th century [4] and their development over the millennium from 500 AD
to 1500 AD represents the outstanding technological development of this period [2]. Early
waterwheels (such as the 16 overshot wheels that formed the large Roman mill of 300 AD at
Barbégal, near Arles, France, and generated perhaps 20 kW [5]) were geared down, so that
the millstones turned more slowly than the waterwheels. This changed as designs improved
over the centuries; by the Middle Ages mills were geared up as much as 5:2 [4]. It is widely
considered [2, 3, 6] that the most dramatic industrial consequences of waterwheels occurred
in the Middle Ages, when the scale of milling increased considerably with the development of
large towns. Their considerable economic and social impact may be judged by the increased
application of waterwheels [3, 7, 8]. From grinding grain and pumping water in antiquity,
water powered mills were developed to forge iron, full cloth, saw wood and stone, and for
metalworking and leather tanning. Later, waterwheels were applied to drive the machines of
the early industrial revolution.
The power of waterwheels increased by a factor of three during the 18th century, to
perhaps 10 kW. Much effort went into the scientific investigation of their efficiency. In 1704
Antoine Parent calculated the maximum efficiency of an idealized undershot waterwheel. In
England John Smeaton (founder of the society of Civil Engineers) made scale models of both
undershot and overshot waterwheel designs, during 1752–4. He varied components to establish
empirically the most effective designs, and concluded that undershot were no more than 22%
efficient whereas overshot were 63% efficient [2, 9]. In 1780 Leonhard Euler studied the
latest waterwheel developments. In the early 19th century the French engineer J V Poncelet
increased the power of undershot waterwheels, as we shall see, to that of overshot wheels [2, 6].
The famous 1835 paper of Coriolis was written, not on the subject of earth rotation, but rather
on energy transfer in rotating systems such as waterwheels [10].
In this paper we shall develop simple models of the two main types of vertical waterwheel,
overshot and undershot, which will bring to light the important technical issues and permit
a realistic calculation of waterwheel efficiency. We shall see why overshot wheels are
more efficient, why undershot continued to be used despite this, and why the modifications
introduced by Poncelet were able to increase drastically undershot waterwheel efficiency. Our
overshot waterwheel model of section 3 is quite general, and shows that waterwheels are stable
dynamical systems even under rapidly changing load. Both overshot and undershot models
are simple, yet illustrate which of the design parameters are important.

2. Idealized waterwheel

We begin with an idealized overshot waterwheel model, which will serve to introduce notation
and some basic concepts. Consider figure 1. Here 12 triangular buckets are attached to the
rim of a wheel. Each bucket is free to swivel about a horizontal axis. The buckets are filled
with water, which here is assumed to drop vertically from a flume. The filled buckets cause the
waterwheel to rotate. At a ‘spill angle’ θ1 near the rim of the wheel is a baffle that causes the
buckets to shed water—no water is spilled otherwise. We assume that the wheel is frictionless,
and does work turning a millstone. The effect of the millstone is represented by a constant load
torque G L . We shall determine the equation of motion for this waterwheel, and shall estimate
its efficiency.
The wheel has been constructed so that torque is applied solely from the gravitational
potential energy of the water, and not from its momentum. The more realistic model in what
The efficiency of overshot and undershot waterwheels 195

θ1

Figure 1. Idealized overshot waterwheel powered by gravitational potential energy (water head).
Water drops vertically into buckets, and remains there until tipped out at angle θ1 .

follows will include water momentum, as well as permit water spillage and allow for friction,
and other such unavoidable realities. Here we can assume that those buckets at angle θ , where
0 < θ
θ1 , are filled with water and that all the others are empty. There are n buckets each
occupying an angle θ around the rim, so that n θ = 2π. The mass of water in each bucket
is m = ρ f t, where ρ is the water density (kg m−3 ), f is the flow rate (m3 s−1 ) and t
is the time interval over which water fills the bucket. We can obtain this from ω t = θ ,
where ω is waterwheel angular speed, so that
ρf
m = θ. (1)
ω
Let us ignore the possibility of water overflowing, due to slow rotation rate, or due to fast
rotation (centrifugal force). The torque applied about the waterwheel axle by the weight of
water is G = m g R sin(θ ) per bucket, where R is the wheel radius and θ is the location
of the filled bucket. For large n we can calculate the total torque as follows:


ρg f R θ1 ρg f R
G≈ dθ sin(θ ) = [1 − cos(θ1 )]. (2)
ω 0 ω
Again assuming that n is large, we shall ignore the torque (opposing that of equation (2))
arising from the buckets being emptied when they reach θ1 . For this idealized waterwheel the
equation of motion is
I ω̇ = G − G L (3)
where I is the moment of inertia of the wheel plus water, and where ω̇ = dω/dt. We shall not
solve (3), but simply note that there is a stable state for this system with angular speed
ρg f R
 = [1 − cos(θ1 )] (4)
GL
as is readily seen from (2) and (3).
We can calculate the efficiency of this waterwheel as follows. The energy input by each
bucketful of water is E in = 2m g R and the corresponding power is Pin = E in  θ1
196 M Denny

Figure 2. Overshot waterwheel with canted vanes (bucket separators). As the cant angle φ
increases, water is retained for longer, increasing torque.

assuming that the wheel rotates at the constant steady-state rate ω̄. The total input power is
θ1
Pin = 2π n Pin = 2ρg f R. Useful output power is Pout =  G L and so waterwheel efficiency
is found to be
 
Pout 2 1
ε≡ = sin θ1 . (5)
Pin 2
So the efficiency of this idealized waterwheel is independent of all the parameters except the
spill angle. This is a consequence partly of our idealizations and partly of our assumption that
the wheel is powered solely by gravity, i.e. by the weight of water alone, and not its momentum.
Efficiency can be 100% if the spill angle is θ1 = π, so that the water contributes to torque
until it is at the bottom of the wheel. For practical waterwheels to which we now turn, this is
difficult to achieve, so efficiency is reduced.

3. General overshot waterwheel efficiency

To make the overshot waterwheel model more realistic, we must make a number of changes.
Firstly, real waterwheels did not have pivoted buckets—this device was adopted above to ensure
that water did not spill out as the wheel turned. Instead, the rim of the wheel is partitioned
off into sections, as suggested in figure 2. These rotate with the wheel, and so water spills out
increasingly as θ increases. Also, water splashes over the sides as it flows in to the buckets,
and flows from one bucket to lower buckets as the wheel turns. To allow for these effects, we
assign a loss factor x(θ ) to the buckets, describing the fraction of water that remains in the
bucket. Thus x(0) = 1 and x(π) = 0; for other values of θ the loss factor takes on values that
depend upon waterwheel design.
Water also may be lost due to centrifugal force. If the bucket walls are along radial lines
(φ = 0, in the notation of figure 2) then water is shed from the wheel for angles exceeding
θmax , for which the centripetal part of the gravitational force exceeds the centrifugal force.
This leads to
ω2 g
cos(θmax ) = , ωmax
2
≡ . (6)
ωmax
2 R
The efficiency of overshot and undershot waterwheels 197

The total mass of water in the buckets, at any given instant, can be found from the foregoing
to be


ρ f θmax ρf
m= dθ x(θ ) ≡ X 0 (ω) (7)
ω 0 ω
where the integrated loss factor X 0 depends upon ω through equation (6). Similarly, from a
calculation analogous to that of the last section, the total torque due to gravity is


ρg f R θmax ρg f R
Gg = dθ x(θ ) sin(θ ) ≡ X 1 (ω). (8)
ω 0 ω
There is an additional driving torque G w , due to the momentum of the water from the headrace
(the flume leading to the waterwheel), determined as follows. The force of water striking the
vanes on the waterwheel rim is F = ρ f (v − ω R), i.e. the mass per second of water striking
the vanes, multiplied by the change in speed of the water as a consequence of striking the
vanes [11]. Here v is the component of water speed that is tangential to the wheel, and ω R is
the speed of the vanes. The torque is thus
G w = ρ f R(v − ω R). (9)
The equation of motion is (see equation (3)) I ω̇ = G g + G w − G L − G k . Here we have allowed
for kinetic friction, for example at the waterwheel axle, generating a torque G k , assumed
constant. Substituting from equations (7)–(9):
 
ρf ρg f R
ηM R 2 + X 0 (ω)R 2 ω̇ = X 1 (ω) + ρ f R(v − ω R) − G L − G k . (10)
ω ω
The wheel has mass M and moment of inertia ηM R 2 , where 12 < η < 1; the precise value of
η depends upon waterwheel structure. Again, G L is the torque due to the load. Equation (10)
is valid for low values of water speed v and low angular speed ω. For higher values of v
we might expect the water to be less effective than suggested in (9) at imparting torque to the
waterwheel2 , while for higher ω we should include a speed-dependent air resistance term. As a
practical limit upon the validity of (10), we note that if v is large enough so that G w = G L + G k
then the steady-state angular speed is ω̄ = ωmax . For larger v there is no steady state. Instead,
angular speed is always increasing. Yet it is observed that, even in the absence of load, a
waterwheel does not accelerate indefinitely. Thus, we consider that (10) is valid if G w
G k ;
for larger values of v we must include other energy-dissipating terms in (10).
The steady-state angular speed ω̄, found by setting the right-hand side of equation (10) to
zero, is the solution to
ρg f R
G L + G k − ρ f R(v − ω̄ R) ≈ X 1 (ω̄). (11)
ω̄
We expect (see what follows) ω̄  ωmax and so we can solve (11) iteratively:
ρg f R X 1 (0) ρg f R X 1 (ω̄(0) )
ω̄(0) = , ω̄(1) = ,.... (12)
G L + G k − ρ f Rv G L + G k − ρ f R(v − ω̄(0) R)
We can see by inspection of (10) that this state is stable: if ω = ω̄ + δ then sign(ω̇) =
−sign(δ). Thus, any perturbation away from ω̄ causes angular speed to revert to ω̄.
In figure 3 we show the result of numerically integrating equation (10) for two cases (see
table 1), differing in the form assumed for the loss factor x(θ ) of equations (7) and (8).
This confirms that the steady-state angular speed is stable. For the numerical integration
we have recast equation (10) in terms of dimensionless variables
ds X 1 (s) − ss̄ X 1 (s̄) − s(s − s̄) ω
= , s≡ , τ = ωmax t,
dτ Cs + X 0 (s) ωmax
ω̄ ηMωmax
s̄ ≡ , C≡ . (13)
ωmax ρf
2 Water that has just struck a vane may be unable to ‘get out of the way’ of water that is about to strike the vane. So,
some of the high-speed water momentum is dissipated before being imparted to the vane.
198 M Denny

0.9

0.8
(I) 10

0.7
(II) 10
0.6

0.5
s
0.4 (I) 100

0.3
(II) 100
0.2

0.1

0
0 2 4 6 8 10 12 14 16 18 20
τ

Figure 3. Dimensionless angular speed s versus dimensionless time τ for the overshot waterwheel
model of equation (13). The dimensionless steady-state angular speed s̄ is set equal to 1 in the
absence of load, and is set to 0.1 when a load is applied at τ = 10 (see equation (12)). The
waterwheel quickly adjusts to the new load, for cases (I) and (II) of table 1, and for parameter
values C = 10 and 100. The waterwheel operates more efficiently with a large load than with a
small one. See section 4 for details.

Table 1. Two loss factors x(θ ) considered in the text, and the corresponding integrated loss factors
X 0,1 .

x(θ ) X 0 (ω) X 1 (ω)

 
ω2 ω2
Case I 1 cos−1 1−
ωmax
2 ωmax
2
  
ω4 1 ω4
Case II cos(θ ) 1− 1− 4
ωmax
4 2 ωmax

There are thus two parameters in this model, C and s̄.

We can obtain a general expression for the dependence of ω̄ upon water speed v, by
differentiating equation (11) with respect to v. If the load G L and frictional torque G k are
independent of v then this yields
   2 
d ω̄ 1 X 1 (ω̄) −1 −1 ω̄ R
= 1 + 2x(θ̄max ) + , θ̄max = cos . (14)
dv R cos(θ̄max ) g
In deriving (14) we have made use of equations (6) and (8). Equation (14) shows that, whatever
the form of the loss factor x(θ ), the steady-state angular speed increases with water speed.
Similarly we can show that ω̄ decreases as wheel radius R increases.
Now we are in a position to calculate the overshot waterwheel efficiency. Compared to the
idealized case we find that the input power changes from Pin = 2ρg f R to 2ρg f R + 12 ρ f v 2 ;
the extra term is the power of the flowing water [11]. Output power is Pout = ω̄G L as before,
and so waterwheel efficiency is
X 1 (ω̄) G L
ε=  . (15)
2 + v GL
2
2g R
The efficiency of overshot and undershot waterwheels 199

The derivation of (15) is valid only if G w
G k , and so we require G L
G L . Let us say for
simplicity that water speed v from the headrace has been chosen so that the torque imparted
by water momentum transfer exactly balances friction at the axle: G w = G k and so G L = G L .
From (15) we see that, from the form of the integrated loss factor X 1 (ω̄), efficiency decreases
as steady-state angular speed increases. Also efficiency falls as water speed v increases (X 1 (ω̄)
decreases as ω̄ increases and so, from (14), as v increases). From a similar argument we see
that efficiency is increased as wheel radius R increases. So efficient design calls for low v,
low ω̄ and large R: this accords with experience and observation of real waterwheels. Thus
we conclude that the simple model developed here captures significant features of overshot
waterwheel physics.
If the vane angle φ of figure 2 is greater than zero, then water is shed from the wheel for
angles exceeding θmax , where
R
cos(θmax − φ) = ω2 cos(φ) (16)
g
(see equation (6)). Repeating the development above for nonzero φ leads to, assuming Case I
for X 1 (ω̄) (more precisely, a design with X 1 (ω̄) → 1 as ω̄ → 0) and small ω̄
1 + sin(φ)
ε= (17)
2 + 2gv R
2

so efficiency is increased if φ > 0. Again, this appears to be the case in practice, since most
overshot waterwheels have buckets which are canted in this way, with the partitions between
buckets (the vanes) as shown in figure 2. If we adopt the reasonable parameter values φ = 30◦ ,
v = 2 m s−1 , and R = 2 m then ε = 71%, which is close to Smeaton’s figure of 63%. From
(15) and (17) we see that the parameters that most influence overshot waterwheel efficiency
are (assuming small ω̄) φ and X 1 (0).

4. Undershot waterwheel efficiency

Here we present a simple model of undershot waterwheels from which we calculate their
efficiency, and show why Poncelet’s modification improved things significantly. Before doing
so, it is appropriate here to discuss the reason why this mattered so much. Given the threefold
superiority of overshot efficiency, why persist with undershot waterwheels at all? To understand
this, we must appreciate the ubiquity of waterwheels in Europe,at the beginning of the industrial
revolution, prior to the widespread availability of steam engines. We have already stressed
the importance of waterwheels since antiquity, as a source of power. The Domesday Book of
1086 recorded over 5000 mills in England. By 1820 France alone had 60 000 waterwheels.
The dense population of mills along early 19th century European rivers and streams meant
few hydro sites, so water head (height difference, and thus potential energy) became a scarce
and valuable resource. Overshot wheels required a large head (2–10 m) and so were usually
confined to hilly areas, or required extensive and expensive auxiliary construction, such as mill
races (water flumes or sluices) that ran for hundreds of metres. Undershot wheels, on the other
hand, could operate with less than 2 m head and so could be located on small streams in flat
areas, nears to population centres. Thus they remained important well beyond the period when
scientific investigation had shown them to be relatively inefficient. The French government
offered large prizes for improved waterwheel design, and this spurred a lot of theoretical and
experimental investigations. Poncelet won a prize with his modified waterwheel vanes, which
proved to be an immediate success.

Conventional undershot design

To estimate the efficiency of undershot waterwheels, consider figure 4(a). We shall simplify
the analysis by assuming that wheel radius is large, so that the water flow is normal to the
200 M Denny

v' = ωR

v
v'

ω
v

v' = 0

Figure 4. (a) Undershot waterwheel with radial vanes. The water flow approaching (receding
from) the wheel has mean speed v(v  ). The wheel rotates at constant angular speed ω̄ = v  /R.
(b) Poncelet’s modification: curved vanes that trap the water, releasing it only when the water has
transferred most of its horizontal momentum. Achieving this requires a careful balancing of vane
shape and water flow rate.

vanes. Thus, if the vane area is A, then the mass of water that presses against each vane per
unit time is ṁ = ρ A(v − v  ). Here v is mean water speed before transferring momentum
to the waterwheel, and v  = ω R ≡ cv is the mean water speed afterwards, both assumed
constant. Thus we expect 0 < c < 1. The force exerted by the water against the vanes is
F = dtd (m(v − v  )) = ρ Av 2 (1 − c)2 . The output power of the waterwheel, resulting from
this force, is Pout = Fv  ; this is the applied force multiplied by the distance moved by the
vanes per unit time [11]. Thus Pout = ρ Av 3 c(1 − c)2 .
The input energy is d E in = 12 ρ Av 2 dx for a water lamina of width dx. So the input power
is Pin = 12 ρ Av 3 , since dtd x = v. Thus, the waterwheel efficiency is
Pout
ε= = 2c(1 − c)2 . (18)
Pin
This peaks for c = 13 (so that the waterwheel vanes move at a third of the initial water
speed in the millrace) so that the maximum efficiency of the undershot waterwheel is about
30%. Given the simplicity of our analysis, this is remarkably close to Smeaton’s figure of
22%. We have not allowed for loss of energy (due, for example, to water splashing) or for
the finite wheel radius, both of which would reduce our estimate. The equation of motion
for the undershot waterwheel is I ω̇ = G w − G L − G k where the torque due to water flow is
G w = F R = ρ A R(v − ω R)2 . A steady-state angular speed ω̄ is found by setting ω̇ = 0, as
before; if the waterwheel design is such that ω̄ R = 13 v then this is the most efficient operating
speed. For such a design the load can be expressed in terms of other parameters:
G L = 49 ρ A Rv 2 − G k . (19)
Thus water speed is the most important factor in determining maximum load, for undershot
waterwheels.
The efficiency of overshot and undershot waterwheels 201

Poncelet modification
This consisted of a careful reshaping of the vanes, as shown in figure 4(b). These curved vanes
hold the water as the wheel rotates; the water falls back (off the outside edge of the vane)
with zero speed, v  ≈ 0, if the vane is properly adjusted to the water speed. Thus efficiency
is improved for two reasons. Firstly, a gravitational component of torque is provided, as
with overshot wheels. Secondly, more of the mill race water momentum is transferred to the
wheel. We can account for this latter effect as follows. The analysis is as for conventional
undershot design except that now the force exerted by water pressing against the vanes is given
by F = ρ Av 2 (1 − c), since here the speed difference of the water, resulting from interaction
with the vane, is approximately v, and not v − v  . Calculating input and output powers as
before leads to the following expression for Poncelet waterwheel efficiency:
ε = 2c(1 − c) (20)
(see equation (18)). We expect that this is an underestimate, since our simple analysis does not
allow for the extra gravitational contribution to torque. From (20) we see that efficiency peaks
for c = 1/2 at ε = 50%. This is a significant improvement. The efficiencies obtained by
Poncelet were higher than this (about 65%); we attribute the difference to gravitational torque,
as described above. The combination of undershot design with some gravitational power input
is known as a breast shot waterwheel, and is a 19th century invention, combining elements
of overshot and undershot designs. The most efficient of these is the Poncelet type. From
figure 4(b) we see that the curved vanes anticipate the shape of hydraulic turbines.

5. Summary and discussion

For pedagogical purposes we have been able to model waterwheel physics by using only
elementary concepts (force, torque, energy, power) from mechanics and fluid dynamics.
The simple overshot and undershot waterwheel models developed here can account
quite well for the measured efficiencies. The undershot model explains why the Poncelet
modification significantly improved efficiency. Both models highlight those parameters that
are important in waterwheel design. For example, it is not difficult to derive the following
expression for the ratio of undershot to overshot power output
Pus 8 v2
≈ . (21)
Pos 27 g R(1 + sin(φ))
(Here we have made a number of assumptions. The undershot wheel is optimum, and the
overshot wheel has X 1 (ω̄) ≈ 1 and low friction.) For realistic parameter values this shows
that the overshot wheel is significantly more powerful, particularly for low flow rate and large
wheel radius.

Acknowledgments

The author is grateful to JSTOR library services of U Michigan for providing the full
reference [9] for Smeaton’s paper. This paper ‘An Experimental Enquiry concerning the
Natural Powers of Wind and Water to Turn Mills, and Other Machines, Depending on a
Circular Motion’ was originally delivered to the Royal Society in May 1759.

References
[1] Strogatz S H 1994 Nonlinear Dynamics and Chaos (Reading, MA: Addison-Wesley) pp 301–11
[2] Waterwheels Encyclopaedia Britannica CD 98
[3] Usher A P 1988 A History of Mechanical Inventions (New York: Dover) chapter 7
[4] Landels J G 1978 Engineering in the Ancient World (London: Constable) chapter 1
202 M Denny

[5] James P and Thorpe N 1994 Ancient Inventions (New York: Ballantine) chapter 9
[6] Mason S F 1962 A History of the Sciences (New York: Macmillan) pp 74, 105, 276
[7] Gies F and Gies J 1995 Cathedral, Forge, and Waterwheel: Technology and Invention in the Middle Ages (New
York: Harper Collins)
[8] Hutchinson Encyclopedia 1998 (Godalming, UK: Helicon) pp 1131–2
[9] Smeaton J 1759 Phil. Trans. R. Soc. 51 100–74
[10] Coriolis G G J 1835 Ec. Polytech. 15 142–4
[11] Douglas J F and Matthews R D 1996 Fluid Mechanics (Harlow, UK: Longman) chapter 8